JSONPath - RFC9535 compliant JSON Path library

Congratulations on the release. Do you plan on implementing the ability to validate JSON Path expressions at compile time?

Edit:
I ask because there are PostgreSQL functions which work with JSON Path and it’d be cool to write Ecto helpers which can, at compile time, tell you “hey, this path is wrong”.

Thank you

The code already checks that the argument is a valid JSON Path expression before doing any evaluation, and returns an error when the expression is invalid. I forgot to include examples both in the README and in the announcement post You can find some examples in the project’s README now for invalid expressions. I also added bang functions in case you want to do @query JSONPath.build!(“…”) .

As for SQL integration, unfortunately the syntax is different for SQL/JSON compared to RFC-9535. For example the query "$[?@.age < 27].name” from the example becomes something like "$[*] ? (@.age < 27).name”. My use case is RFC-9535 expressions exclusively, I have no plans at the moment of doing anything with SQL/JSON queries + Ecto.

New version 0.3 released

JSONPath.evaluate/3 now takes an optional 3rd argument to specify the returned type. Possible values are:

• :values - Returns the node values, same as previous versions
• :paths - Returns the normalized paths
• :values_and_paths - Returns a list of 2-element tuples {node_value, normalized_path}

Examples:

root = %{"people" => [%{"name" => "Alice", "age" => 20}, %{"name" => "Bob", "age" => 30}]}
query = "$.people[?@.age > 25]"
JSONPath.evaluate(root, query, :values)
# {:ok, [%{"name" => "Bob", "age" => 30}]}

JSONPath.evaluate(root, query, :paths)
# {:ok, ["$['people'][1]"]}

JSONPath.evaluate(root, query, :values_and_paths)
#{:ok, [{%{"name" => "Bob", "age" => 30}, "$['people'][1]"}]}

That’s a great feature but I have a question. Do you think encoding the different result formats in a single struct makes sense? Changing the return type through a flag argument is a smell in my opinion.

Yes I wasn’t fully convinced either. I might change it to 3 separate functions instead in a future version. I also take suggestions if you have a better idea, before I make another questionable decision